So some Gemini prompts use far more energy than this: Dean gives the instance of feeding dozens of books into Gemini and asking it to provide an in depth synopsis of their content. “That’s the form of thing that may probably take more energy than the median prompt,” Dean says. Using a reasoning model could even have the next associated energy demand because these models take more steps before producing a solution.
This report was also strictly limited to text prompts, so it doesn’t represent what’s needed to generate a picture or a video. (Other analyses, including one in ’s Power Hungry series earlier this yr, show that these tasks can require far more energy.)
The report also finds that the overall energy used to field a Gemini query has fallen dramatically over time. The median Gemini prompt used 33 times more energy in May 2024 than it did in May 2025, in keeping with Google. The corporate points to advancements in its models and other software optimizations for the improvements.
Google also estimates the greenhouse gas emissions related to the median prompt, which they put at 0.03 grams of carbon dioxide. To get to this number, the corporate multiplied the overall energy used to reply to a prompt by the typical emissions per unit of electricity.
Somewhat than using an emissions estimate based on the US grid average, or the typical of the grids where Google operates, the corporate as a substitute uses a market-based estimate, which takes under consideration electricity purchases that the corporate makes from clean energy projects. The corporate has signed agreements to purchase over 22 gigawatts of power from sources including solar, wind, geothermal, and advanced nuclear projects since 2010. Due to those purchases, Google’s emissions per unit of electricity on paper are roughly one-third of those on the typical grid where it operates.
AI data centers also devour water for cooling, and Google estimates that every prompt consumes 0.26 milliliters of water, or about five drops.
The goal of this work was to offer users a window into the energy use of their interactions with AI, Dean says.
“Individuals are using [AI tools] for every kind of things, they usually shouldn’t have major concerns in regards to the energy usage or the water usage of Gemini models, because in our actual measurements, what we were capable of show was that it’s actually reminiscent of stuff you do without even occupied with it every day,” he says, “like watching just a few seconds of TV or consuming five drops of water.”
